'Caterpillar Ballot' cost Kerry more than 1000 Cuyahoga votes

[AirAmFan]

Conservative Estimate: Kerry's vote margin tumbled by more than a thousand votes in Cuyahoga alone because of a scrambled, "caterpillar" ballot" in overcrowded polling places.

The Congressional debate over the challenge to Ohio's electoral votes repeatedly cited hours-long lines at the polls in pro-Kerry areas as the main source of unfair election results. Senator Barbara Boxer estimated that, in Columbus alone, from 5,000 to 10,000 people left the polls in frustration without having voted. Hundreds of thousands of Ohioans who had intended to vote doubtless never got that far, having heard about long lines or seen them on TV.This thread reports on a subtler consequence of the long lines. In Cuyahoga alone, over a thousand voters waited for their turn to vote, punched out the chad next to Kerry's name, but had their vote counted for George Bush or for one of the minor candidates! Across Ohio, this same phenomenon may have robbed Kerry of many additional thousands of votes.

A generation ago, political science research established that having one's name listed first on a ballot gives a candidate an advantage of up to several percentage points, especially in minor contests where few voters recognize any of the candidates' names. A 1976 addition to the Ohio Constitution ordains that"The general assembly shall provide by law the means by which ballots shall give each candidate's name reasonably equal position by rotation or other comparable methods to the extent practical and appropriate to the voting procedure used." (From http://www.legislature.state.oh.us/constitution.cfm?Part=5&ExpandsSections=Yes -- thanks to DUer "Niche" for this URL.)

But this directive has been carried far past the point of sanity. There are 1436 precincts in Cuyahoga, but only 584 polling places. At 447 of these polling places, from 2 to 10 different precincts vote together.

"Ballot rotation" at the PRECINCT LEVEL, not the POLLING PLACE level, sets a trap for unwary voters when the location is crowded and chaotic. I call this trap the "Caterpillar Ballot", because, at the same polling place, a valid vote for John Kerry (or any other candidate) might have to go in ANY OF FIVE DIFFERENT LINES ON THE BALLOT. Kerry's name crawls from ballot line to ballot line, just like a caterpillar crawling from twig to twig on a tree. The Cleveland Plain Dealer explained it all in an article that remains obscure (it is or was at http://www.cleveland.com/news/plaindealer/index.ssf?/base/cuyahoga/1102674912293811.xml ). Here's my explanation, which may or may not be clearer to you:

Let's say Precinct 31 has Kerry on line 1, Precinct 32 has Kerry on line 2, Precinct 33 has Kerry on line 3, Precinct 34 has Kerry on line 4, and Precinct 35 has kerry on line 35. Let's say that all five of these precincts are assigned to vote together in one huge overcrowded school gymnasium. Everything will be fine as long as voters from each precinct use only the punch machines set up for that precinct's candidate order. In well-staffed suburban locations in Cuyahoga, apparently is is common for a staffer to accompany a voter to the right booth, insert the ballot into the machine for her, and draw the curtain closed.

But at crowded and chaotic polling places, which tended to be located in predominantly pro-Kerry areas in 2004, voters often are left on their own to negotiate hours-long lines waiting for a booth. First-time and other inexperienced voters may not know it makes a difference which machine they use to punch out their chads. Seeing a fairly short line, a Precinct 31 voter might take her ballot to a Precinct 34 machine. She'd punch a hole where it says "Kerry", on Line 4. But a stamp on the back of her ballot would route it to a counting machine where a Kerry vote must be on Line 1. If she took it to another precinct's machine, she'd cast a Kerry vote on another line, but still not on line 1 where it would have to be, to be counted for Kerry.

Do you see the problem? Kerry's name crawls like a caterpillar from one ballot line to another, depending on which machine is used in a room full of machines set up for different precincts.Just like the "Butterfly Ballot" that made Palm Beach County famous in the 2000 Florida fiasco, the Ohio Caterpillar Ballot has great potential for mischief. And it is the most straghtforward explanation I've encountered for incredibly large vote totals minor candidates racked up in some Cuyahoga precincts.

In subsequent posts here, I'll lay out (1) my 1000-net-vote conservative estimate of the effect of "caterpillar ballot crawl" on John Kerry's race for President.

Then (2) I'll explain the methodology. A formula I've derived under reasonable statistical assumptions gives the data an equal chance to say either major candidate lost votes to "caterpillar crawl". The key factors are (3) the proportion of the major candidate vote that went to Kerry, and (4) unexpectedly high vote proportions for minor candidates. The candidate that most people intend to vote for loses the most votes when conditions are chaotic.

Miscounted votes have to go somewhere else, and they don't all go to the other major candidate. Excessive vote proportions for minor candidates are the most frequent tell-tale signs of "caterpillar crawl". Another wrinkle that to an extent limits the ability to reconstruct miscounts entirely reliably is (5) the failure of Cuyahoga's ballot counting program even to tally votes for the "Disqualified" candidate, Ralph Nader. Rather than simply remove Nader's name from the ballot, many Ohio counties had a line in the Presidential election for "Disqualified Candidate". But, at least in Cuyahoga, the ballot counting machines were programmed not to even tally votes on the "Disqualified" line. That made it a "black hole" for at least 140 Kerry votes in Cuyahoga alone.

[DoYouEverWonder]

1. Funny how all of these glitches and problems

continue to always favor Bush.

[AirAmFan]

24. This work could pinpoint the location of PHYSICAL EVIDENCE of vote

miscounting, depending on how ballots are stored after an election, and for how long they are kept intact!

This work, and work that other people are doing with publicly-available data for other Ohio counties ,identifies precincts where significant numbers of ballots were miscounted. The 33 polling places listed in post #13 together comprise 79 precincts.

If it still is possible physically to inspect all the ballots that were used to generate each line of the precinct-by-precinct "Canvass Report", then a hand recount of those specific precincts could include inspection of the BACK of all ballots. Certainly for Benedictine High's precincts Cleveland 4F (#1806) and Cleveland 4N (#1814) (see post #5), and very likely for all of the precincts identified here, such inspection would find more than a thousand Cuyahoga ballots that were counted for one precinct, but are stamped on the back in blue with the name of another precinct. And less conservative definitions of excessive votes for minor party candidates, plus recovery of precinct tallies for "Disqualified Candidate", would lead to recovery of hundreds or thousands of additional votes in Cuyahoga alone. (See the notes at the end of post #6.)

Statewide application of the methodology developed here, or similar methodologies, could lead to recovery of thousands to tens of thousands of miscounted votes.

[L. Coyote]

26. Random cross-voting favors the candidate with the least support.

However, things were not random in Cuyahoga.

Ballot orders are discussed in detail in the 2004 Ohio Election - Presidential Ballot Orders and Cross-Voting page. It discusses all the ballot order combinations and the probabilities of vote-switching for the possible combinations.

Another Web page at the site 2004 Ohio Election - Analysis, Summary, Charts, and Spreadsheets and the article "How Kerry Votes were Switched to Bush Votes!" discusses specific examples in Cuyahoga County and some of the non-random aspects of cross-voting.

There are links to spreadsheets with results of the candidates and non-votes and statistical analysis summaries. One spreadsheet subdivides the county according to the numbers of precincts and ballot orders at locations. The Web page discussions show how non-random their distribution is, factors which must be accounted for in calculations of the number of switched votes.

[AirAmFan]

27. Your assertions are not intuitive. Could you outline the insights on which

they are based in heuristic terms, or give a specific short example of what you mean?

It sounds as though you find what I'm saying here clear enough to understand, and also it sounds like you disagree. Where exactly do you think I went wrong?

[Goldeneye]

2. 1000 votes here, 1000 votes there

here a vote, their a vote, everywhere a stolen vote, old Mr. Blackwell is a scam.....e...i...e...i...o

[sustarr]

3. lol!!

goldeneye that is hilarious!!

[L. Coyote]

28. there should be a few OINK OINK's in there somewhere! n/t

[okieinpain]

4. oh man, this is such bull shit. sorry for cursing, but there is no

other word for this. this is just ridiculous, why are we just hearing about this, what the fuck is kerry/dnc/dlc doing. this is just pissing me off to no end.

I'm truly thinking about going to the green party, the dem party is either just to damn stupid at the leadership level, or they have been hanging out with the same people that armstong williams has.

[AirAmFan]

5. URL links for the data I used, and an egregious example

All the data that went into my 1000 net-vote estimate came from the Cuyahoga County Board of Elections website. The best way to explain exactly what these data represent is with a good example.

The Benedictine High School polling place in Cleveland (listed on page 11 of http://boe.cuyahogacounty.us/boe/PDF/votinglocations.pdf ) is the precinct cluster where the "caterpillar" did the most damage to Kerry. This precinct has been discussed everywhere the "ballot rotation" issue has been on the agenda, including the December 10 Plain Dealer article I linked in the lead-in.

About 2 percent of the multi-megabyte "Canvass Report" at http://boe.cuyahogacounty.us/boe/results has the Presidential election results for Cuyahoga. The Benedictine High location, in a heavily pro-Kerry area on Martin Luther King Drive in Cleveland, was where two different precincts voted, with two different ballot orders.

The results (Precinct Name(Precinct ID number), Badnarik votes, Bush, Kerry, Peroutka) were

Cleveland 4F (1806) 0 20 299 215
Cleveland 4N (1806) 164 12 334 10

Note that Badnarik got 164 of his 1886 countywide votes at Cleveland 4N, but none at Cleveland 4F. And Peroutka got 215 of his 1751 countywide votes at 4F, but only 10 at 4N. What could explain these very strange results.

The answer very likely is frustrated attempts to vote for Kerry on the "wrong" machine. Note that 95 percent of the major party vote at Benedictine went to Kerry (no big surprise on MLK Drive). You still can access images of the actual ballot overlays for the punch machines used at Benedictine, at http://boe.cuyahogacounty.us/boe/ballots/pdf/CLEVE04F.PDF and http://boe.cuyahogacounty.us/boe/ballots/pdf/CLEVE04N.PDF . You'll see that the ballot orders were:

Precinct 4F: Kerry Disqualified Peroutka Badnarik Bush
Precinct 4N: Badnarik Bush Kerry Disqualified Peroutka

This observed pattern of minor-candidate votes is exactly the what you'd expect to see if substantial numbers of people trying to vote for Kerry in both precincts used the other precinct's punch machine. Kerry was on line 1 for 4F, and Peroutka was on line 3. And Kerry was on line 3 for 4N, with Badnarik on line 1.

Each voter had to go to the right precinct to have her name found on the reigstration list and get a punchcard stamped on the back in blue with the name of the precinct. If 200 or so 4F voters tried to vote for Kerry on one of 4N's machines, their votes would have been miscounted for Peroutka. Not Badnarik, not Bush, and not Disqualified, but Peroutka. And if 160 or so 4N voters had punched Kerry's chad on one of 4F's machines, their votes would have been misdirected to Badnarik. Not Peroutka, not Disqualified, and not Bush. Since the data do not reject our ballot order explanation, and nobody has supplied a plausible alternative, we can be pretty confident that the "Ohio Caterpillar Ballot" robbed John Kerry of more than 300 votes at this one polling place. And there were 446 other clusters of precincts where something similar could have happened.
My conservative estimate (given in detail in a future post) is that at least 32 other Cuyahoga precincts had the same problem as Benedictine.

[AirAmFan]

22. Oops! Precinct number for Cleveland 4N is 1814, not 1806.

[Time for change]

50. This is becoming clearer and clearer

I think we can figure this out now -- see my post # 29.

[AirAmFan]

6. (1) Summary of results of estimating miscounts and their effect on the vote

Conservative estimates of effect of "caterpillar ballot
crawl" in Cuyahoga County Ohio, Presidential race,
General Election 2004

1033 Net Kerry gain from restoring miscounted votes (KNET)

1061 Gross gain to Kerry from restoring miscounts   (KGAIN)
  28 Gross gain to Bush from restoring miscounts    (WGAIN)

 143 "Black hole" votes intended for Kerry         
(MDK)
  53 "Black hole" votes intended for Bush          
(MDW)

1285 Total gross miscounts in 33 polling places     (TOTMISC)

 334 Excess votes for Badnarik                      (XB)
 545 Excess votes for Peroutka                      (XP)

SOURCE: Calculations based on data from URLs linked in post #5
at
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=172x5139

NOTES:  "Black hole" votes are miscounts credited
to "Disqualified Candidate", for which the
Cuyahoga BOE furnished and tallied no results.
To supply these missing tallies, the BOE would
have to reprogram its counting machines and
re-run all ballots through them.

"Excess votes" for minor candidates are defined
here as at least 3 times the countywide average
after adjusting for outliers like Benedictine
High.  Where BC is the cluster vote for Badnarik,
PC is the cluster vote for Peroutka, and TOTVOTC
is the total cluster vote for all four
candidates, XB = max(0,BC-0.0066*TOTVOTC, and
XP = max(0,PC-0.0063*TOTVOTC).  These very
generous tolerances within which minor-candidate
votes are considered believable and intended are
the driving force that make the overall estimates
conservative.  Less generous tolerances would
bring more polling places into the list of
miscount locations and drive the effect on
Kerry's margin higher.

Jan 10 2005 8:30pm
Author: AirAmFan

[AirAmFan]

7. (2a) Methodology. I posted my first stab at a statistical theory for the

caterpillar ballot about a month ago, in a thread archived at http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=203&topic_id=147718&mesg_id=147718 . I've simplified the notation and corrected some errors born of haste, but the theory I'm using still is basically the same. I'm going to try to make this thread stand on its own for most readers, by being more heuristic than formal.

As shown more formally in my archived thread, the total number of miscounted votes in a multiprecinct cluster depends on the following factors:

c = "Cluster size", the number of precincts voting together in the same place. (In Cuyahoga, c ranged from 1 to 10).

xB = Number of excess votes for Badnarik (See the Notes in post #6)

xP = Number of excess votes for Peroutka.

m(K) = the proportion of major-candidate votes in the cluster that went to Kerry`

gBK = Number of votes Badnarik gains from every c*(c-1) wrong-machine votes intended for Kerry.
gPK = Number of votes Peroutka gains at Kerry's expense.

gBW = Number of votes Badnarik gains from every c*(c-1) wrong-machine votes intended for Bush.
gPW = Number of votes Peroutka gains at Bush's expense.

The last four variables are entirely theoretical, flowing from just the number and configuration of ballot orders at a precinct. The attached table of "Wrong-machine beneficiaries" gives the values of these variables for Cuyahoga: Here's a quick heuristic explanation:

With two precincts at one location (call them X and Y), there are two ways a voter who wants to mark a ballot for Kerry can go wrong: A voter registered at X could use a Y machine, or a voter registered at Y could use an X machine.

With three precincts at one location (call them X, Y, and Z), there are six ways a voter who wants to mark a ballot for Kerry can go wrong: A voter registered at X could use a Y or Z machine, or a voter registered at Y could use an X or Z machine, or a voter registered at Z could use an X or Y machine. In general, with c precincts at one location, there are c*(c-1) ways a Kerry voter could go wrong. There also are c*(c-1) ways a Bush voter could go wrong.

As seen for Benedictine High, once you know the configuration of ballot orders for all the precincts at a location, it is straightforward (though tedious) to enumerate all the ways a voter could go wrong. Let's assume that all the ways a Kerry voter could go wrong are equally likely, and all the ways a Bush voter could go wrong also are equally likely. Then it is worthwhile to enumerate all the possibilities for wrong voting, and to tally the number of votes out of each c*(c-1) intended for a major candidate that would wind up benefiting each other candidate. I wrote a computer program to do this for all the ballot configurations that occurred in Cuyahoga in November 2004. The results are in the attached "Wrong-machine beneficiaries" table. Note "B" stands for Badnarik, "W" for Bush, "K" for Kerry, "D" for Disqualified, and "P" for Peroutka. Note also that, both in the columns that tally wrong Kerry vote beneficiaries and in the columns that tally wrong Bush-vote beneficiaries,
the number of votes always sums to c*(c-1). In notation, where gXY stands for the expected gain to candidate X at candidate Y's expense when a voter uses a wrong machine,

gBK + gWK + gKK + gDK + gPK = c*(c-1), and
gBW + gWW + gKW + gDW + gPW = c*(c-1).

The table has a case number in column 1, the cluster size ("c") in column 2, and then the 10 "gXY" parameters in columns 3 through 12. Column 13 gives the number of different ballot-order configurations corresponding to the case, column 14 the total number of polling places having the indicated theoretical pattern, and the final column lists all the configurations explicitly.

For instance, the Benedictine High location has c=2 with Badnarik at the top of one ballot order and Kerry at the top of the other. This is configuaration "BK" of case 4 in the table, one of 65 locations with a pattern where Badnarik and Peroutka benefit from miscounted votes at Kerry's expense. The table adds a detail we didn't discuss before: Disqualified (the "Black Hole" and Peroutka benefit from miscounted votes at Bush's expense for this case. We didn't need to talk about it at Benedictine High, because 95 percent of the major party vote went to Kerry. The formula used for the full Benedictine calculation is balanced and takes into account possible miscounting of the five percent of Benedictine votes intended for Bush. At Benedictine, it was the data, not the formula, that tilted the estimate of miscounted votes to the Kerry side.

The "Wrong machine beneficiaries" table is a tremendous reduction of a complex problem involving 447 locations with 96 ballot order configurations down to 28 basic cases (note 7a and 7b have the same ratios of all the gXYs to each other).

[AirAmFan]

9. delete

[AirAmFan]

10. (2b) "Wrong-machine beneficiaries" table for Cuyahoga Presidential election

Who would have gained had a major-candidate voter used another
precinct's voting machine in error?
                                       Number
         Wrong-machine vote meant for: of
        ------------------------------ poll-    Num-
             Kerry     |      Bush     ing-     ber
        -------------- |-------------- place    of
         Possible      | Possible      ballot-  poll-
Case     vote gainers  | vote gainers  order    ing
num-    -------------- |-------------- configu- pla-
Ballot-order
ber   C  B  W  K  D  P | B  W  K  D  P rations  ces 
configurations
---   - -- -- -- -- -- |-- -- -- -- -- -------- ----
-----------------
                       |
 A.  Caterpillar crawl impossible or makes no difference
                       |
 1.   1  -  -  -  -  - | -  -  -  -  -    5     137  B,W,K,D,P
                       |
 2.   2  0  0  2  0  0 | 0  2  0  0  0    5      20 
BB,WW,KK,DD,PP
 3.   3  0  0  6  0  0 | 0  6  0  0  0    2       2  BBB,DDD

B.  Observe minor-candidates credited with votes meant for
both Bush
    and Kerry
                       |
 4.   2  1  0  0  0  1 | 0  0  0  1  1    5      65 
BD,BK,KP,WD,WP
                       |
 5.   3  1  2  0  2  1 | 2  0  2  1  1    5      64 
BDP,BWK,BWP,KDP,
                       |                              WKD
 6.   3  2  1  0  1  2 | 1  0  1  2  2    5      43 
BKD,BKP,BWD,WDP,
                       |                              WKP
 7a.  4  3  3  0  3  3 | 3  0  3  3  3    5      27 
BKDP,BWDP,BWKD,
                       |                             
BWKP,WKDP
 7b.  5  5  5  0  5  5 | 5  0  5  5  5    1       4  BWKDP

 8.   3  2  0  2  0  2 | 0  2  0  2  2    8      26 
BBD,BBK,BKK,KKP,
                       |                             
WDD,WPP,WWD,WWP
                       |
 9.   5  4  5  2  5  4 | 5  2  5  4  4    7      14 
BBWDP,BBWKP,BKDPP,
                       |                            
BWDPP,BWWKD,WKDDP,
                       |                             WKKDP
                       |
10.   4  2  3  2  3  2 | 3  2  3  2  2    5       8 
BBDP,BWPP,BWWP,
                       |                             
KDPP,WWKD
11.   4  3  2  2  2  3 | 2  2  2  3  3    6       7 
BBKP,BKDD,BKKD,
                       |                             
WDPP,WKKP,WWKP
12.   5  5  4  2  4  5 | 4  2  4  5  5    4       7 
BWKDD,BWKKP,
                       |                            
BWKPP,WKDPP
                       |
13.   4  1  4  2  4  1 | 4  2  4  1  1    4       6 
BBWP,BWWK,KDDP,
                       |                              WKKD
14.   5  4  4  4  4  4 | 4  4  4  4  4    3       4 
BBDDP,BKKDD,WWKDD
                       |
15.   4  4  1  2  1  4 | 1  2  1  4  4    2       3  BBKD,BKKP
                       |
16.   6  6  7  4  7  6 | 7  4  7  6  6    3       3 
BBKDPP,BBWWKD,WWKKDP
17.   6  7  7  2  7  7 | 7  2  7  7  7    2       3 
BWKDDP,BWKKDP
                       |
18.   4  4  0  4  0  4 | 0  4  0  4  4    2       2  BBDD,WWPP
                       |
19.   5  1  6  6  6  1 | 6  6  6  1  1    1       1  BDPPP
20.   5  2  6  4  6  2 | 6  4  6  2  2    1       1  WKKDD
21.   6  6  5  8  5  6 | 5  8  5  6  6    1       1  BWWWPP
                       |
22.   7  4 12 10 12  4 |12 10 12  4  4    1       1  WWKKKDD
23.   9 11 18 14 18 11 |18 14 18 11 11    1       1  BBWWWKKKD
24.  10 20 20 10 20 20 |20 10 20 20 20    1       1 
BBWWKKDDPP

No Kerry votes to minor candidates
Observe minor-candidates credited with votes meant for Bush
                       |
25.   2  0  1  0  1  0 | 1  0  1  0  0    5     114 
BP,BW,DP,KD,WK
                       |
26.   3  0  2  2  2  0 | 2  2  2  0  0    8      15 
BBP,BPP,BWW,DDP,
                       |                             
DPP,KDD,WKK,WWK
27.   4  0  4  4  4  0 | 4  4  4  0  0    2       3  BBPP,WWKK
                       |
28.   4  0  3  6  3  0 | 3  6  3  0  0    1       1  DPPP
                                       -----   -----
                            TOTALS:      96     584

SOURCE: Calculations from Cuyahoga County Board of Elections
data.

NOTES:  "C" stands for "cluster size", the
number of precincts voting together
at one location.  "B" stands for candidate Bednarik
and, in the final column,
for ballot order (Bednarik, Bush, Kerry, Disqualified, and
Petroukis).  "W"
stands for candidate Bush and for ballot order (Bush, Kerry,
Disqualified,
Petroukis, Bednarik).

  "K" stands for candidate Kerry and for ballot
order (Kerry, Disqualified,
Petroukis, Bednarik, Bush).
  "D" stands for "Disqualified candidate"
and for ballot order (Disqualified,
Petroukis, Bednarik, Bush, Kerry).
  "P" stands for candidate Petroukis and for ballot
order (Petroukis, Bednarik,
Bush, Kerry, Disqualified).

[AirAmFan]

8. (2c) Methodology for cases 4 through 24

(2c) Methodology (Continued)

Careful examination of the 28 cases of "caterpillar crawl" in the table shows there are three basic situations for numerical estimation of miscounts.

First, consider "easy" cases 1 through 3. In 159 of 584 polling places, it was impossible to use a "wrong" machine or it would have made no difference had a voter used a "wrong" machine. These are the 137 single precinct polling places, with configurations B, W, K, D, or P; the 20 polling places with patterns BB, WW, KK, DD, or PP; and the two polling places with patterns BBB or DDD. In these cases, there was only one ballot order used at each location. It would have made no difference had a voter from one precinct used a punch machine set up for another precinct. If the Cuyahoga BOE were interpreting the state ballot rotation law sensibly, all polling places would fall into this category.

Second, for cases 4 through 24, there are 292 polling places where theoretical patterns of "caterpillar crawl", both for voters intending to vote for Kerry and for voters intending to vote for Bush, eventuate in miscounts in favor of minor candidates

Third, there are 133 polling places where no theoretical patterns of caterpillar crawl for votes intended for Kerry eventuate in miscounts favoring minor candidates. In these configurations (including most notably BP, BW, DP, KD, and WK), any wrong votes intended for Kerry would benefit either Bush or Disqualified (the "Black Hole"). Fortunately, in all these 133 polling places, thoretical patterns of caterpillar crawl for votes intended for Bush all include a minor candidate among their beneficiaries. By making slightly stronger statistical assumptions, and by exploiting symmetries among the gXYs, results still can be obtained for these cases. But they are on shakier ground than results among the 292 polling places for "Calculation Class B" (4 through 24), as exlained below under "Calculation Class C"

Thus two formulas need to be developed, one for the 133 "tougher" cases just mentioned and one for the 292 mentioned just before them ("Calculation Class B")

Calculation Class B

The table gives the split among the candidates for wrong-machine miscounts, but what about the scale? An effective estimation formula would use all available data that are relevant, give them equal weight, and not be rendered useless by the unavailability of data on Disqualified.

If all balloters intended to vote for Kerry. The theoretical sum of (gBK + gPK) would have an observable sample analog in (XB + XP). Multiplying by c*(c-1)/(gBK + gPK) would give it the correct scale for the total cluster miscount.

Similarly, if all balloters intended to vote for Bush, the theoretical sum of (gBW + gPW) would have an observable sample analog in (XB + XP). Multiplying by c*(c-1)/(gBW + gPW) would give it the correct scale for the total cluster miscount.

The appropriate multiplier is somwhere between the two special cases of universal intent to vote for Kerry and universal intent to vote for Bush. Using the observed cluster proportions of major candidate votes to measure overall voter intent as a probablility and taking the expected value gives

Total miscount

= c*(c-1)*{xB+xP}*{(m(K)/(gBK + gPK)) + ((1-m(K))(gBW + gPW))}

There are two denominators in this formula. For calculation class B, neither denominator is zero, and so this formula may be applied whenever we have reason to believe "caterpillar crawl" may have occurred with the configuations listed for cases 4 through 24.

[AirAmFan]

45. Clarification: In the next revision of my paper, I'll add some language

to this section to make it absolutely clear what exactly the "total miscount" for a cluster of precincts represents. After the heading "Calsulation Class B", I will insert:

We cannot observe transfers of votes between major candidates DIRECTLY, but we can use the information in the table (post #10) to estimate them indirectly.

For major candidates, who have spent millions of dollars to get voters to recognize their names and to appeal to voters' hopes and fears, it's very difficult to estimate the number of "excess" votes that might arise from miscounting rather than from effective psychological manipulation of voters' emotions. But excess votes for minor-party candidates, who've been able to afford little publicity, are much easier to estimate. Diagnostic outlier models identify polling places where minor candidates far exceeded their minuscule average countywide vote proportions, about one half of one percent for Badnarik and Peroutka combined. (See posts numbers 18 and 36).

The table shows the theoretical ratio of hard-to-observe vote shifts between major candidates to easier-to-estimate vote shifts from major-party candidates to minor-party candidates. For example, in polling places whose configurations of ballot orders make them part of Case 5 in the table, shifts from Kerry to Bush (2 of every 6 miscounted Kerry votes) should equal the total of shifts from Kerry to Badnarik and from Kerry to Peroutka. And shifts from Bush to Kerry (2 of every 6 miscounted Bush votes) should equal the total of shifts from Bush to Badnarik and from Bush to Peroutka. Exactly the same can be said about shifts of votes from major-party candidates to Disqualified, which we could have observed directly had the BOE tallied them.

The total miscount in a polling place is a weighted average of all these vote shifts, using Kerry's proportion of the site's major candidate vote to weight shifts away from Kerry, and using Bush's proportion to weight shifts away from Bush. From the parameters for Case 5, we know that multiplying the estimated number of major-party-to-minor-party vote shifts by three = (6 / (1 + 1)) is a clever way to turn estimates of vote shifts we can see into estimates of total vote shifts, including those we do not observe directly.

The table gives the split of miscounted votes, but what about the scale, for the general case of c precincts per cluster rather than just the special case c=3 in Case 5? ...

[AirAmFan]

11. (2d) Methodology for remaining cases 25 through 28 of the table ('Class C')

In all cases in Cuyahoga, theoretical patterns of expected gainers at the expense of the major candidates include one or more of Badnarik, Peroutka, or Disqualified. Had the BOE provided tallies of votes cast for Disqualified, the single formula

Total miscount

= c*(c-1)*{xB+xP+DC}*{(m(K)/(gBK + gPK + gDK)) + ((1-m(K))(gBW + gPW + gDW))}

(where the new variable DC represents the number of votes for Disqualified in the cluster) would have sufficed for all configurations of ballot orders in all cases. This formula would have applied the assumptions of equiprobable Kerry voter miscounts and equiprobable Bush voter miscounts by simply adding the three observable gain coefficients at each major candidate's expense together, without weighting any of the three more heavily than the other two. And there never would have been zeros in denominators to worry about.

Just as excess votes for Badnarik or Peroutka brought sites like Benedictine High into the list of sites where "caterpillar crawl" is presumed to have occurred, excessive votes for "Disqualified" would have flagged sites where votes were being lost to major candidates by sloppy BOE procedures. Without actual data on votes for Disqualified, these sites mainly will stay hidden.

But we might make a little bit of progress with stronger statistical assumptions and use of symmetries apparent in the "Wrong-Machine Beneficiaries" table. Instead of the two separate assumptions we've made so far--equiprobable Kerry voter miscounts and equiprobable Bush voter miscounts--let's make the stronger assumption of equiprobable miscounts for all who intend to vote for a major candidate. This assumption would be unjustified if, for example, Bush voters tended to be older than Kerry voters and to have more experience in voting. In this event, Bush voters would be more likely to know it might make a difference which voting booth they used to punch out chads, and the line of thought we're about to pursue would be invalid.

In any case, a formula for cases 25 through 28 of the table ("Calculation Class C" is on shakier ground and less reliable than the formula used for cases 4 through 24 ("Calculation Class B"). In a way, we're like the motorist who lost car keys in a dark parking lot, and looked for them under lampposts because the light was better, not in the dark places where the keys were more likely to be. But let's press on.

Compare the columns of five parameters for Kerry voters with the columns of five parameters for Bush voters. In a given line of the table, you'll see the same individual numbers on the left as you see on the right. For example, the parameters for case 9 are 4-5-2-5-4 for Kerry voters and 5-2-5-4-4 for Bush voters. You never see more than three different values for the gXY parameters in any case. And there are patterns of left-right, left-left, and right-right symmetry in the parameters:

gDK = gKW = gWK = gBW, gKK = gWW, and gPK = gDW = gBK = gPW

in every line of the table.

Then the assumption of equiprobable miscounts for all major-candidate voters means that (gDK / gBW) have the ratio (m(K) / (1-m(K)). This suggests using a sample analog of this expression, (DC / gBW) = (m(K) / (1-m(K)), to substitute for the unobserved DC in the general formula.

In cases 25 through 28 in the table, it always is true that gBK = gPK = gPW = 0, but that gDK and gBW are positive. Substitution and rearrangement of the general formula above gives us a second- or third-best formula for these cases:

Total miscount

= c*(c-1)*{xB+xP}*{(1-m(K))/(gBW + gPW + gDW)}

+ c*(c-1)*((m(K))**2)/{gDK*(1-m(K))}

[AirAmFan]

12. (2e) Factors driving miscount estimates for individual clusters

                                    Minor can-     Black
                                    didate ex-     Hole
  Config-        Cluster            cess votes     votes
  ura- ---------------------------  ------------  ----------
c tion Page Sequence     Name        XB     XP    MDK   MDW

  Cases 4 through 24 ('Calculation Class B')

2 BD     24  263 ALEXIA_MANOR_____   0.40   0.00  0.00  0.02
2 BD     39  426 NORTH_OLMSTED_BRA   0.17   0.00  0.00  0.07
2 BD     15  164 WADE_PARK_APARTME   0.00   3.06  0.00  0.04

2 BK     11  120 BENEDICTINE_HIGH_ 157.04 218.36  0.00  9.46
2 BK     24  266 JOHN_F._KENNEDY_H   0.15   0.00  0.00  0.00
2 BK     21  226 RIVERVIEW_COMMUNI   0.00   0.27  0.00  0.01

2 BK     21  232 WALTON_ELEMENTARY   0.00   0.06  0.00  0.01
2 BK     24  262 WILLOW_ELEMENTARY   2.46  16.53  0.00  1.78
2 KP     52  567 CHRISTIAN_REF_CH_   0.00   1.70  0.00  0.04

2 KP     28  311 GATEWAY_MANOR____   6.86  94.00  0.00 21.80
2 KP      4   43 LAWRENCE_SCHOOL__   0.00   1.02  0.00  0.42
2 KP     12  124 MT_HAVEN_BAPTIST_  37.82  66.15  0.00  1.71

2 KP     52  568 WARRENSVILLE_HTS.   0.00   1.55  0.00  0.05
2 WD     21  235 ERNEST_BOHN_TOWER   0.00   2.42  0.00  0.07
2 WD     21  230 PILGRIM_CONGREGAT   2.81   0.00  0.00  0.32

2 WD     13  136 ROBERT_H._JAMISON   0.00   2.98  0.00  0.05
2 WD     13  143 UNION_ELEMENTARY_   0.00   0.89  0.00  0.06
2 WP     25  277 DOUGLAS_MACARTHUR   0.00   2.39  0.00  0.63

2 WP     22  241 MARION_STERLING_E   0.00   5.33  0.00  0.12
2 WP     23  254 ROBERT_FULTON_ELE   0.00   4.05  0.00  0.05
3 BBK    27  296 NEW_CHAMBERS_SCHO  30.90   0.00  0.00  0.44

3 BDP    25  271 EMILE_E._DESAUZE_   0.00   1.23  1.17  0.02
3 KDP    48  529 SOLON_CENTER_FOR_   0.00   1.60  0.79  0.28
3 WDP    11  115 HARVEY_RICE_ELEME   0.00   1.32  0.32  0.02

3 WKD    16  171 CORY_UNITED_METHO  44.52  24.82 61.99  2.73
3 WKD    19  204 HOLY_REDEEMER_SCH  20.52   6.96 24.01  1.29
3 WKP    21  231 LAKEVIEW_TOWERS_A   6.61  42.76 11.45  2.62

4 BKDP   41  453 TOWN_HALL________   4.39  19.92  8.03  4.12
6 BBKDPP 40  436 ROYAL_REDEEMER_LU   0.00  20.08  5.34  5.05

  Cases 25 through 28 ('Calculation Class C')

2 BP     19  210 BROOKLAWN_ELEMENT   4.40   0.00  6.92  0.00
2 BP     31  337 GARFIELD_HTS_MIDD   0.10   0.00  0.10  0.00
2 BP     34  373 GRANT_SCHOOL_____   0.00   0.20  0.04  0.00
2 BP     18  191 Y._M._C._A.______  15.06   5.23 22.76  0.00

Notes: 'Page' refers to the BOE PDF listing polling places at
http://boe.cuyahogacounty.us/boe/PDF/votinglocations.pdf .
"Sequence" is the sequential position of the
polling place in this same document, running
from 1 on page 1 to 583 on page 53.  The other
columns are explained in the totals for these
results, provided in post #6 at
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=172x5139

Jan 11 2005
Author: AirAmFan

[AirAmFan]

13. (2f) Bottom-line miscount results for individual clusters

  Config-        Cluster
  ura- ---------------------------   Total
c tion Page Sequence     Name     Miscount  KNET  KGAIN  WGAIN

  Cases 4 through 24 ('Calculation Class B')

2 BD     24  263 ALEXIA_MANOR_____   0.46   0.34   0.40   0.06
2 BD     39  426 NORTH_OLMSTED_BRA   0.27  -0.03   0.12   0.15
2 BD     15  164 WADE_PARK_APARTME   3.16   2.97   3.07   0.09

2 BK     11  120 BENEDICTINE_HIGH_ 393.47 355.60 374.53  18.93
2 BK     24  266 JOHN_F._KENNEDY_H   0.17   0.15   0.16   0.01
2 BK     21  226 RIVERVIEW_COMMUNI   0.30   0.24   0.27   0.03

2 BK     21  232 WALTON_ELEMENTARY   0.09   0.04   0.06   0.02
2 BK     24  262 WILLOW_ELEMENTARY  22.08  14.93  18.51   3.58
2 KP     52  567 CHRISTIAN_REF_CH_   1.80   1.61   1.70   0.10

2 KP     28  311 GATEWAY_MANOR____ 133.77  46.53  90.15  43.62
2 KP      4   43 LAWRENCE_SCHOOL__   1.57  -0.13   0.72   0.85
2 KP     12  124 MT_HAVEN_BAPTIST_ 107.30 100.44 103.87   3.43

2 KP     52  568 WARRENSVILLE_HTS.   1.66   1.44   1.55   0.11
2 WD     21  235 ERNEST_BOHN_TOWER   2.56   2.27   2.42   0.15
2 WD     21  230 PILGRIM_CONGREGAT   3.36   2.07   2.71   0.65

2 WD     13  136 ROBERT_H._JAMISON   3.08   2.88   2.98   0.10
2 WD     13  143 UNION_ELEMENTARY_   1.01   0.76   0.88   0.13
2 WP     25  277 DOUGLAS_MACARTHUR   3.32   0.77   2.05   1.27

2 WP     22  241 MARION_STERLING_E   5.57   5.08   5.33   0.24
2 WP     23  254 ROBERT_FULTON_ELE   4.16   3.94   4.05   0.11
3 BBK    27  296 NEW_CHAMBERS_SCHO  47.64  29.99  30.88   0.89

3 BDP    25  271 EMILE_E._DESAUZE_   3.66   4.48   3.46  -1.02
3 KDP    48  529 SOLON_CENTER_FOR_   4.13   0.88   1.82   0.94
3 WDP    11  115 HARVEY_RICE_ELEME   2.01   2.20   1.93  -0.26

3 WKD    16  171 CORY_UNITED_METHO 202.40 226.04 180.49 -45.55
3 WKD    19  204 HOLY_REDEEMER_SCH  79.79  85.71  69.45 -16.26
3 WKP    21  231 LAKEVIEW_TOWERS_A  76.61  71.00  67.42  -3.58

4 BKDP   41  453 TOWN_HALL________  48.64  19.58  28.03   8.45
6 BBKDPP 40  436 ROYAL_REDEEMER_LU  48.18  -2.64  13.93  16.58

  Cases 25 through 28 ('Calculation Class C')

2 BP     19  210 BROOKLAWN_ELEMENT  19.12  12.83  11.20  -1.64
2 BP     31  337 GARFIELD_HTS_MIDD   0.32   0.14   0.15   0.01
2 BP     34  373 GRANT_SCHOOL_____   0.13   0.07   0.07  -0.00
2 BP     18  191 Y._M._C._A.______  64.13  40.41  36.24  -4.17

Notes: 'Page' refers to the BOE PDF listing polling places at
http://boe.cuyahogacounty.us/boe/PDF/votinglocations.pdf .
"Sequence" is the sequential position of the
polling place in this same document, running
from 1 on page 1 to 583 on page 53.  The other
columns are explained in the totals for these
results, provided in post #6 at
http://www.democraticunderground.com/discuss/duboard.php?az=view_all&address=172x5139

Jan 11 2005
Author: AirAmFan

[skids]

14. Question: How are the other races rotated?

Since you seem to have looked at the ballot orders pretty thoroughly, are the rotations used on lower ticket races associated with the rotations for the U.S. President race? That is, were there a limited number of ballots, such that you can tell the order of the lower ticket race from the order of the presidential race?

I'm wondering both from the perspective of determining bias in the ballot rotations, and from the perspective of using 3rd party votes on lower ticket races to help improve the estimate?

Another question -- has anyone looked to see if this happened in 2000 or previous years?

[AirAmFan]

17. Excellent questions! 'I don't know' is my answer to both

I've focused exclusively on the Presidential ballot rotation. As far as I know, there was only one version of the ballot for each precinct, and I've downloaded PDF images of all 1436 of them using a script. I haven't looked beyond page 2 of each ballot image, where the Presidential race is listed, but I saved the PDFs to CDR so that I could do exactly what you're suggesting about the minor races sometime in the future, probably after the BOE has wiped its online directories for 2004. I would surmise that there must be some set of algorithms giving for each state and local race the precinct ballot order corresponding to having each of the 5 Presidential candidates (B, W, K, D, and P) at the top of the Presidential ballot page. I should think that in at least some state and local races in some precincts, excessive votes for minor party candidates would go hand-in-hand with excessive minor-party votes for President. But if such races are far down in the list of about 50 contests, ballot roll-off might limit the usefulness of such data. It's definitely something I plan to look at but I've given it lower priority than other tasks.

As for your second question: the history of ballot rotation in Cuyahoga and throughout Ohio is the biggest gap in my knowledge. I don't know whether the insane practice of rotating ballots by PRECINCT, rather than by POLLING PLACE, dates from an era preceding the clustering of most precincts and has not been revisited for partisan reasons, or whether it's something that was instituted recently.

[skids]

19. Depending on the PDF...

I could probably extract the ballots into some more managable format. That is, if they are not just scans.

If you have a list of URLs handy, I could go fetch a second "backup copy" and have a look.

[AirAmFan]

21. Give me a couple of days and I'll post something here, as long as you

intend to share your ballot order resource in TXT or XLS with the rest of us.

Each URL is the same, except for an 8-letter abbreviation for the precinct name just ahead of the ".PDF". For two examples, see post #5 about the two precincts that voted together at Benedictine High. In one or two posts right here, with HTML turned off, I could supply you a file with the 1436 precinct numbers from the Canvass Report and the corresponding 8-letter precinct name abbreviations, three to a line without wrapping. For example, the Benedictine precincts would look like:

1806 CLEVE04F 1814 CLEVE04N

Would that be satisfactory? Will you share any ballot-order resource resulting from your work with the rest of us?

[skids]

23. That sounds good.

Yeah, the base URL and the list of abbreviations would be fine. Of course I'd share the results.

[AirAmFan]

25. I've just posted these data for all but 450 (of 1436) precincts at URL

http://www.democraticunderground.com/discuss/duboard.php?az=show_topic&forum=172&topic_id=5180 . If all goes as planned, I'll post the third and last chunk before tomorrow. If you have any questions on these data, please post a reply on THAT thread, not this one.

[skids]

20. Oh, BTW, here is what you are dealing with for rolloff.

Unsure how "preliminary" this data was, but I happened to have it on hand. Looks like it bottoms out before 50%, though I don't have the state/local partisans in this table.

                        booth   absentee
Fingerhut-D		303085	38590
Kerry-D	        	448486	52943
Total cast		602048	85212
Connally-J		303995	36862
Fruerst-J		339817	40137
O'Donnel-J		269782	29046
Moyer-J	        	205075	25258
Lanzinger-J		173215	20613
O'Neill-J		198992	28809
Pfeifer-J-U		341056	40941
Bush-R	        	221606	29088
Voinovich-R		334204	39777

[AirAmFan]

15. (3) Kerry's major-candidate vote proportion & excess minor-party votes

Clusters with excessive minor-candidate votes, sorted by
Kerry's proportion of major candidate votes

                                               Cluster
                                     
-------------------------
 KPRO    KNET      XB      XPc CONFIG Pg.  Seq.     Name

0.9731   3.94   0.000   4.0532 WP     23   254
ROBERT_FULTON_EL
0.9721  29.99  30.902   0.0003 BBK    27   296
NEW_CHAMBERS_SCH
0.9707   2.97   0.000   3.0692 BD     15   164
WADE_PARK_APARTM

0.9685   2.20   0.000   1.3253 WDP    11   115
HARVEY_RICE_ELEM
0.9680 100.44  37.826  66.1522 KP     12   124
MT_HAVEN_BAPTIST
0.9670   2.88   0.000   2.9852 WD     13   136
ROBERT_H._JAMISO

0.9613   0.15   0.159   0.0002 BK     24   266
JOHN_F._KENNEDY_
0.9594   4.48   0.000   1.2373 BDP    25   271
EMILE_E._DESAUZE
0.9560   5.08   0.000   5.3362 WP     22   241
MARION_STERLING_

0.9519 355.60 157.044 218.3602 BK     11   120
BENEDICTINE_HIGH
0.9459   1.61   0.000   1.7082 KP     52   567
CHRISTIAN_REF_CH
0.9429   2.27   0.000   2.4252 WD     21   235
ERNEST_BOHN_TOWE

0.9340   1.44   0.000   1.5592 KP     52   568
WARRENSVILLE_HTS
0.9188 226.04  44.525  24.8203 WKD    16   171
CORY_UNITED_METH
0.9081   0.24   0.000   0.2742 BK     21   226
RIVERVIEW_COMMUN

0.9028  85.71  20.529   6.9603 WKD    19   204
HOLY_REDEEMER_SC
0.8972  71.00   6.614  42.7683 WKP    21   231
LAKEVIEW_TOWERS_
0.8755   0.76   0.000   0.8972 WD     13   143
UNION_ELEMENTARY

0.8737   0.34   0.406   0.0002 BD     24   263
ALEXIA_MANOR____
0.8381  14.93   2.469  16.5382 BK     24   262
WILLOW_ELEMENTAR
0.8077   2.07   2.816   0.0002 WD     21   230
PILGRIM_CONGREGA

0.7346   0.04   0.000   0.0692 BK     21   232
WALTON_ELEMENTAR
0.7237  12.83   4.403   0.0002 BP     19   210
BROOKLAWN_ELEMEN
0.7100  40.41  15.060   5.2392 BP     18   191
Y._M._C._A._____

0.6795   0.07   0.000   0.2062 BP     34   373
GRANT_SCHOOL____
0.6739  46.53   6.865  94.0082 KP     28   311
GATEWAY_MANOR___
0.6610  19.58   4.397  19.9254 BKDP   41   453
TOWN_HALL_______

0.6471   0.14   0.103   0.0002 BP     31   337
GARFIELD_HTS_MID
0.6166   0.77   0.000   2.3992 WP     25   277
DOUGLAS_MACARTHU
0.5799   0.88   0.000   1.6003 KDP    48   529
SOLON_CENTER_FOR

0.4751  -2.64   0.000  20.0816 BBKDPP 40   436
ROYAL_REDEEMER_L
0.4595  -0.13   0.000   1.0212 KP      4    43
LAWRENCE_SCHOOL_
0.4463  -0.03   0.175   0.0002 BD     39   426
NORTH_OLMSTED_BR

Jan 11 2005
Author: AirAmFan

[AirAmFan]

16. 'Caterpillar ballot crawl' would have no effect on the election if both major

candidates were equally po;ular in the sites where it occurred. In the estimates posted here, the only polling places that enter into the miscount calculation are those with excessive votes for Badnarik and or Peroutka. In looking for these locations, no attention was paid to the distribution of major-party votes.

But the table in post #15 shows that it turned out that excess minor-party votes were highly correlated with Kerry's popularity. Of 33 locations that entered into the miscount calculation, only 3 (the last three lines of the table) went for Bush, and those not by much.

Notice that net effects on Kerry's margin compared to Bush's of restoring miscounted votes (KNET) at those three sites are negative. And KNET is small in absolutre value at those locations, because the major-candidate split in each case (KPRO) still is close to 50-50.

[L. Coyote]

29. This is incorrect. There is never a collocation with equal probability.

You are overlooking the most significant aspect of cross-voting, switching one major candidate's vote to the other major candidate. This does not create a non-vote or a 3rd party vote trail.

If there were only two candidates and two precincts at a location, cross-voting would switch votes both ways. But, in the 2004 presidential race in Ohio, this is never the case. The precinct probabilities are never equal from the opposing candidate perspectives.

Switching major candidate votes has double the impact of a vote lost to a third party candidate. It subtracts 1 from one column and adds 1 to the opposing column, in effect two votes for one!!

When two precincts have Kerry and Bush collocated in the same position on the ballot orders, one precinct has one set of probabilities that a Kerry vote gets switched to a Bush votes, the next precinct a different set of probabilities, etc. And, in each precinct, the Bush vote switch probabilities are different that the Kerry probabilities.

[AirAmFan]

31. You're not making sense. I've provided a table (in post #10) that

enumerates, for all 101 configurations of ballot orders at Cuyahoga polling places, all possible misvotes at "wrong" machines. This table shows who gains and who loses when Kerry supporters use wrong machines and when Bush supporters use wrong machines. In all cases, vote switches between major candidates are accompanied by vote switches from major candidates to minor candidates or Disqualified. Do you not understand the table or did you not look at it? I'm sorry if the table is not clear.

[L. Coyote]

39. The important cross-votes are the ones that switch Kerry votes to Bush.

You are focused on the votes that switch to 3rd party candidates or non-votes, switches that do not have the same impact as major party switches. When a Kerry vote is switched to a Bush vote, the margin changes by 2 votes, not 1. Major party cross-votes have twice the impact on the results. And there is reason to believe that they could be far more numerous, even though they are not evidenced in any real sense.

By starting with the premise you use, you have looked in all the wrong places. There are high 3rd party vote counts where Kerry votes DO NOT get switched to Bush votes. Why don't you look at the locations where Kerry votes DO get switched to Bush. You are giving the impression that there were fewer cross-votes than is actually the case. I get the impression you are working for Bush, trying to hide the real problem.

What you are doing is showing what rate of cross-voting occurred from major to minor candidates in some locations only.

In Cuyahoga county there are 352 precincts in the 2\2 subset, at locations with 2 ballot orders and 2 precincts. Of these, 220 are crawl 1 and crawl 4 ballot collocations, only 130 are crawl 2 or crawl 3. At the crawl 1 and crawl 4 locations, in one precinct at each location, Kerry cross-votes switch to Bush votes. These precincts are where the greatest impact on the results occurs. There are 110 precincts where the probability of a Kerry cross-vote counts as a Bush vote.

These are the P = 1.0 precincts. Of the 110, 30 are in precincts with >90% Kerry support, 28 have 70-90% Kerry support, 22 have 60-70% Kerry support, 12 have 50-60%, 12 have 40-50, and 5 have <40%. Every single Kerry cross-vote in these precincts counted as a Bush vote.

Do you understand this? If not, read the Web pages linked in my first post in this thread.

Yes, the charts are unclear. But that is not the problem. Where's the chart for the number of precincts where a Kerry cross-vote becomes a Bush vote? Those are precincts without inflated 3rd party numbers because cross-votes don't go to the 3rd parties, they go to Bush!!

[AirAmFan]

44. Huh? I'm "working for Bush, trying to hide the real problem?!"

[AirAmFan]

18. (4) Detecting excessive votes for minor-party candidates

Formulas presented above (in posts #8 and #11) for estimating the total number of miscounts at a polling place depend crucially on the values of measures of excessive votes for minor-party candidates. As explained in a note at the bottom of post #6, the measures used in the estimates presented here are

XB = max(0,BC-0.0066*TOTVOTC, and
XP = max(0,PC-0.0063*TOTVOTC).

Excess votes" for minor candidates are defined here as those above and beyond at least 3 times the countywide average after adjusting for outliers like Benedictine High.

Removing the influence of outliers did not change the magnitude of countywide averages very much, because they were minuscule to begin with. The Cuyahoga "Canvass Report" online at http://boe.cuyahogacounty.us/boe/results has

1,886 0.28 01 = MICHAEL BADNARIK
221,606 32.89 02 = GEORGE W. BUSH
0 04 = CANDIDATE DISQUALIFIED
448,486 66.57 03 = JOHN F. KERRY
1,751 0.26 05 = MICHAEL A. PEROUTKA
-----------
673,729

I sorted clusters of physical precincts listed in the "Canvass Report" by the proportion of votes for Badnarik. Benedictine High was first, with 164 of his 1,886 countywide votes. To find a breakpoint, I looked for dramatic increases in the increment in the proportion as I looked down the sorted list. Then I totaled up the Badnarik votes at those locations, subtracted him from the 1886 Canvass Report total as probable miscounted votes for major candidates, and divided by 673,729. The answer was 0.0022, not very different from Badnarik's original 0.0028.

A similar procedure yielded 0.0021 for Peroutka, not very different from his original 0.0026.

In future work, I plan to experiment with more sophisticated outlier models to see how robust the 1000 net-vote estimate is to alternative specifications.

[L. Coyote]

30. An easy way to determine the 3rd party vote rate is by using all the

precincts with zero probability of a major candidate cross-vote to that candidate. Of course, there are underlying assumptions, like that ballots are counted in the correct precinct, etc.

[AirAmFan]

32. Why don't you start another thread to show what you think is a more

appropriate set of calculations, rather just snipe in ways I don't find clear? If you have SHORT specific examples of ways your methodology produces different results from mine for particular locations, we might be able to communicate about them. But you're using an esoteric jargon here that I do not comprehend. Have you ever posted a DU thread to explain your methodology in clear English? Posts that simply say, "I disagree" are not very helpful.

[L. Coyote]

41. Because you are proposing ideas here that need correction.

The jargon I'm using is part of the Website about cross-voting and the same jargon we have been using in all the previous threads on this issue. We both have been participating in those other threads, so the jargon should be familiar. I assume that simple statistical jargon, like "probability," is understood.

This discussion is about your methodology. You placed it in a discussion forum. Now it is subject to scrutiny. You should welcome that.

I'm pointing to flaws in your logic. I think your conclusions are not valid.

Why are you ignoring the precincts where Kerry cross-votes become Bush votes? Are you trying to cover-up the most significant part of the cross-voting? Why?

Answer these simple question and the discussion can move forward.

[AirAmFan]

43. The good news is that your uninformed 'critique' is becoming more

specific. I can answer specific points, such as "ignoring the precincts where Kerry cross-votes became Bush votes". The bad news is, I still doubt that you've actually read all my posts in this thread, or if you've read them you didn't understand them and didn't ask "clarifying questions" about them. See message #31.

I am NOT "ignoring the precincts where Kerry cross-votes became Bush votes".. Again, if you've looked at the key table I posted in message #10 and it didn't make sense to you, I apologize for my lack of clarity. In my notation, gWK is the number of possible ways (out of a total of c*(c-1), where c is the number of precincts co-located in one polling place) that a voter intending Kerry can punch Kerry's chad on a wrong machine and have her vote credited to Bush instead. The third column of the table in post #10 is the value of gWK for each of the 101 different configurations of ballot orders in Cuyahoga this election. The configurations are grouped into 29 "cases" that have the same values of the gXYs. If I can convince you to finally take a look at the table, notice that column 3 is greater than zero in many of the cases, including numbers 5, 6, 7a, and 7b.

Let's look at the first configuration in case 5. Configuration BDP means that there are three precincts in a cluster, with (B)adnarik at the top of the ballot in one precinct, (D)isqualified listed first in another, and (P)eroutka listed first in the third precinct. Did you analyze this case on your website, where nobody can ask a question? How would you describe this case in your esoteric jargon, which I cannot decipher?

I would say that, of the 6 ways voters intending Kerry can have their votes go awry on wrong machines, Bush and Disqualified each gain in two, while Badnarik and Peroutka each gain in one. The BOE provided us no data on Disqualified, but my assumptions of equiprobable wrong votes for Kerry supporters makes the sum of excess Badnarik votes and excess Peroutka votes at a Case 5 site an estimate of the number of votes switched from Kerry to Bush out of every 6 cast, assuming everyone at that polling place intended to vote for Kerry.

I take as the proportion of voters who intended Kerry at that site the proportion Kerry took of the major candidate vote, and scale back the estimated number of votes switched from Kerry to Bush in that proportion.

Now look at the parameters on the other side of the vertical line for case #5 in post #10. If all voters at the site intended Bush, they show who would gain how many votes at Bush's expense of every 6 that succumbed to caterpillar crawl. Note that Kerry and Badnarik each would gain two.

Do you see that, for Case 5 polling places, if the major party vote split 50-50 between Bush and Kerry, gains to Bush from wrong Kerry votes would exactly cancel gains to Kerry from wrong Bush votes in my formula (post #8)?

But the location won't even get into the list of likely "caterpillar crawl" sites unless either XB or XP is positive.

[AirAmFan]

36. (4a) Alternative outlier model using the interquartile range

A definition of outliers that once was common uses the "interquartile range"--that is, the difference between the 75th percentile and the 25th percentile. Under this definition, an outlier exceeds the 75th percentile by at least 1.5 times the interquartile range. I just calculated quantiles and interquartile ranges for Badnarik's and Peroutka's proportions of the vote at the 584 polling places in Cuyahoga and re-ran my computer program with the alternative defintion of "excessive" minor-party votes. Changes in the results presented in posts #6 and post #13 are very minor, because the new definition yields points of demarcation for "excessive" Badnarik and Peroutka proportions that are very close to the initial definitions. Under the new definition,

XBiqr = excessive Badnarik vote using the interquartile range definition) = max(0,BC-0.00734), and
XPiqr = max(0,BP-0.0649).

Compare these new cutoff points to those listed in post #18 and you'll see the two pairs of definitions are quite close.

[L. Coyote]

40. The interquartile range is typically used for normal distributions.

In the case of cross-votes to third-party candidates, the distribution is not normal. The vote is skewed to begin with and skewed to an extreme by the cross-votes in one direction only.

For the vote county-wide:
The skew for Badnarik's is 22.43.
The skew for Peroutka is 18.79.
The skew for Kerry is 0.113

A normal distribution has no skew.

Peroutka's 0.50 trimmed mean is 0.084 and the second quartile is 0.0.
Badnarik's 0.50 trimmed mean is 0.136 and the second quartile is 0.182.

What is the Badnarik vote in locations where the major candidates do not collocate with Badnarik in the ballot orders?

What is the Peroutka vote in locations where the major candidates do not collocate with Peroutka in the ballot orders?

[AirAmFan]

42. Oops! Forgot to turn emoticons off, and omitted the TOTVOT factor

The equations should read,

XBiqr = max(0,BC-0.00734*TOTVOT) and XPiqr = max(0,PC-0.00649*TOTVOT)

[MercutioATC]

33. Do you live in Ohio?

I ask because, in Ohio, you sign in at a desk according to your precinct. The polling machines are clearly marked with precinct numbers.

You have to know your precinct number to sign in. From there, you just go to a machine with that number on it.

It's not nearly as confusing as you're seeming to imply.

[AirAmFan]

34. Thanks for your reaction. No, I don't live in Ohio, but I've used punch-.

cards to vote in Illinois. Do you think all the thousands of Ohio precincts in all 88 counties have the clear procedures and mistake-proof signage your voting places evidently have had? Three related problems with voting in this country are that Election-Day resources depend on county wealth, there are more than 3000 counties, and resources are not distributed fairly even within counties. Elections in this country are financed in basically the same way public schools are financed, with resulting similar inequity and unfairness for poor people.

Doubtless confusion of the kind I'm suggesting is relatively rare in Ohio pollling places, but so are hours-long voting lines extending outdoors into a steady rain, and so are classrooms with leaky roofs and broken windows. I'm suggesting here that lack of Election-Day resources and confusion over ballot rotations sometimes go hand in hand.

If what I'm suggesting really is impossible, what's your explanation for Peroutka's picking up 215 of his 1751 countywide total otal at one Benedictine High precinct (where Badnirik got none) and Badnarik's picking up 164 of his 1886 countywide vote at the other Benedictine High precinct (which gave only 10 to Peroutka)? See post #5--I think "Caterpillar ballot crawl" is the only explanation that makes sense. And other locations in highly pro-Kerry areas have similarly anomalous results for minor candidates that fit the explanation I'm providing here SUPERBLY.

[MercutioATC]

35. I only know about my polling place, but it had 6 or 7 precincts in it.

There didn't seem to be any confusion.

I didn't say that what you're proposing is IMPOSSIBLE, I just didn't see any evidence of confusing procedures at my polling place.

[AirAmFan]

37. So we're in agreement. Remember, I'm suggesting here that confusion led

to miscounts at just 33 out of 584 Cuyahoga polling places, comprising only 79 of 1436 precincts. That means even in Cuyahoga the overwhelming majority of polling places likely were "clean" and fairly orderly.

Have you been an Ohio voter long enough to know how long polling places have been set up with different machines for different precincts under the same roof? It would make so much more sense to rotate ballots by POLLING PLACE, rather than by PRECINCT. That way, the particular problem I've written about here could not happen anywhere.

[MercutioATC]

38. Forever, as far as I know.

In rural areas, there may only be 1 or 2 precincts per polling place. In more populated areas, more.

I'm not sure there was enough confusion to matter at all, but I agree that standardizing ballots by polling place instead of precinct would be a good idea.

[AirAmFan]

46. Re: 'Switching of votes through ballot order rotation': The 1,000 votes you

mention matches my very conservative estimate, derived in detail in an Ohio Forum thread nine months ago, at http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=172&topic_id=5139&mesg_id=5139 .

'Caterpillar Ballot' cost Kerry more than 1000 Cuyahoga votes

In Cuyahoga alone, over a thousand voters waited for their turn to vote, punched out the chad next to Kerry's name, but had their vote counted for George Bush or for one of the minor candidates! Across Ohio, this same phenomenon may have robbed Kerry of many additional thousands of votes....

A generation ago, political science research established that having one's name listed first on a ballot gives a candidate an advantage of up to several percentage points, especially in minor contests where few voters recognize any of the candidates' names. A 1976 addition to the Ohio Constitution ordains that"The general assembly shall provide by law the means by which ballots shall give each candidate's name reasonably equal position by rotation or other comparable methods to the extent practical and appropriate to the voting procedure used."... But this directive has been carried far past the point of sanity.

There are 1436 precincts in Cuyahoga, but only 584 polling places. At 447 of these polling places, from 2 to 10 different precincts vote together.

"Ballot rotation" at the PRECINCT LEVEL, not the POLLING PLACE level, sets a trap for unwary voters when the location is crowded and chaotic. I call this trap the "Caterpillar Ballot", because, at the same polling place, a valid vote for John Kerry (or any other candidate) might have to go in ANY OF FIVE DIFFERENT LINES ON THE BALLOT. Kerry's name crawls from ballot line to ballot line, just like a caterpillar crawling from twig to twig on a tree. ...

[AirAmFan]

47. (Too late to edit) Wrong thread. But the thread where this belonged is

a wonderful summary of Ohio voting irregularities, 9 months after this thread was posted. See http://www.democraticunderground.com/discuss/duboard.php?az=show_mesg&forum=104&topic_id=4998616&mesg_id=4998616 .

[WCGreen]

48. That is just not true....

The polling places that have more than one or two precincts, have the precinct voting macines clearly marked, seprated and the check in tables for the voting is done at a table designated by precicnt...

Is there the chance for cross over...

Yes....

But that chance is over blown in this report....

I was on the Cuyahoga County Board of elections, not durring this election, and we sent out inspectors to deal with just such problems....

I find it hard to believe it would be that much of a problem....

[StudentOfDarrow]

49. Wow.

This is incredible. When are Americans going to wake up and realize that America is rigged?!?